Technical Report No. 249 - Abstract

Gian Diego Tipaldi, Kai O. Arras
FLIRT -- Interest Regions for 2D Range Data

Local image features are used for a wide range of applications in computer vision, range imaging and robotics. While there is a great variety of approaches for both, detectors, descriptors and combinations thereof for image data and 3D point clouds, there is no method readily available for 2D range data. For this reason, this paper first proposes a set of benchmark experiments based on known data sets for robot navigation to compare detector repeatability and descriptor matching performance. Secondly, the paper introduces FLIRT that stands for fast laser interest region transform, a multi-scale interest region operator for 2D range data. The transform combines a detector based on a curve approximation of the range signal with a descriptor that makes use of occupancy information in a polar region around the detected point. FLIRT combines the best detector with the best descriptor, experimentally found in a comprehensive analysis of alternative detectors and descriptor approaches. We compare the proposed method with three other multi-scale detectors on the direct range signal and a variant of the shape context descriptor. The experiments lead to very similar repeatability and matching performance values that have been found for local image features in the computer vision literature, encouraging a wide range of applications of FLIRT for 2D range data.